Face, hands and upper body, read live in your browser — plus an on-device expression estimate. Nothing is uploaded.